Modeling Self Assembling Peptides Using MELD

Stephen Jones, Alberto Perez

University of Florida, University of Florida

We explore the potential of computational tools to study peptides as a source of biomaterials through self-assembly. This can offer advantages in the design of more stable scaffold structures and improved functional motifs. Despite routine characterization of the 3D assemblies at the fiber level, there is a lack of atomistic details about the assembly scaffold. Computational tools offer unique potential, both to determine these atomic level details and identify novel sequences, but limitations in force fields and inefficient sampling have restricted studies. To overcome these limitations, we employ MELD (Modeling Employing Limited Data) to achieve full atomistic peptide self-assembly in timescales not possible with conventional or replica molecular dynamics. This enhanced sampling method allows peptides to alternate between high energy states capable of overcoming a rugged energy landscape and low energy states where they are funneled into stable assemblies, driven by contact data. We then compare our results to other machine learning tools and find that they are not yet appropriate to investigate peptide self-assembly.